JP2001518668A - Method and system for pre-retrieving information - Google Patents

Abstract

A method and system for pre-fetching data for a client associated with a proxy server hierarchy and data transfer to the client by an enhanced content server. A method for prefetching data identifies a data access pattern and prefetches selection information based on a dynamic interpretation of the data access pattern. The content server or proxy server identifies a client data access reference pattern associated with the content server or proxy server hierarchy. The data access pattern is preferably stored in a statistics table. The content server generates prefetch hint information based on the input from the statistics table. The prefetch hint information annotates the requested information or object requested by the client. As the request information passes through the proxy server hierarchy, it is best to update the prefetch hint information. The pre-fetch hint information annotates the request information requested by the client. The final decision to pre-fetch selection information for the client is based on pre-fetch hint information and pre-fetch hint values. The prefetch hint value is calculated based on local node specific factors such as storage location constraints and access times.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD OF THE INVENTION

The present invention relates to a method and apparatus for dynamically prefetching information or objects for a client, the client comprising, for example, a world wide
It is preferably associated with a hierarchy of proxy servers, such as being able to participate in pre-fetching web objects on the web.

Glossary Internet: A network of networks and gateways that use the TCP / IP set of protocols.

[0003] Client: A client is a computer that issues instructions to a server that performs tasks associated with the instructions.

Server: A computer is a server that executes a task according to instructions from another computer. A web server typically supports one or more clients.

World Wide Web (WWW or Web): Connecting (clicking) a highlighted word or phrase of interest to users searching for information on the Internet, connecting from server to server and from database to database Internet application to switch between. The Internet WWW server supports the client and provides information to the client. The web, which can be thought of as the Internet, where all resources are addressed as URLs,
Use HTML to display information corresponding to a URL and provide point-and-click to other URLs.

[0006] Universal Resource Locator (URL): A URL is a method for uniquely identifying or addressing information on the Internet and can be thought of as a web document version of an email address. The URL can be accessed by a hyperlink. An example of URL identification is "http://www.philipyu.com:80
/table.html ". The URL has four components. Looking at the left side of this example, the first component specifies the protocol to use (http in this case),
It is separated from the rest of this locator by a ":". The next component is the host name or IP address of the target host, and this component has a double slash "//
, And the right side is delimited by a single oblique line “/” (or optionally “:”). The port number is an optional component, and is separated from the host name by “:” on the left side and by a single oblique line “/” on the right side. The fourth component is the actual file name or program name (table.html in this example) and further separates directories and subdirectories using single diagonal lines in a conventional manner. In this example, the extension ".html" means that this is an HTML file.

Hypertext Markup Language (HTML): HTML is the language used by web servers to create and connect to documents viewed by web clients. HTML generates hypertext documents.

[0008] Hypertext Transfer Protocol (HTTP): HTTP is an example of a stateless protocol in which each request from a client to a server is processed independently. The server has no record of previous connections. "Http:" at the beginning of the URL
Indicates that the requesting client and the target server communicate using the HTTP protocol for the specified resource.

Internet Browser or Web Browser: A graphical browser that runs an Internet protocol such as HTTP and displays the results on the user's screen.
Interface tools. The browser can act as an Internet tour guide, complete with image desktops, directories, and search tools for when the user "surfs" the Internet. In the present application, a web browser is a client server that communicates with the World Wide Web.

[0010] Client cache: The client cache is typically used as a primary group (cache) of objects accessed by the client. In a WWW environment, the client cache is typically implemented by a web browser and stores the objects accessed during the current invocation, ie, may be a non-persistent cache, or may store the objects across all invocations. Can be.

[0011] Caching proxy: A specialized server in a network that acts as an agent for a client to look up objects and possibly return a cached copy. The caching proxy is typically invoked as a result of a cache miss from the client cache, and thus generally serves as a secondary or higher level cache.

Meta-tagging: the association of information with arbitrary objects. For example, in HTTP, information can be associated with both requests and responses in fields of the HTTP header. For example, an HTTP client can specify the URL that requested the returned page.

[0013]

[Prior art]

As the popularity of the World Wide Web (WWW or Web) increases,
Internet traffic has increased significantly. The Web is now one of the major bottlenecks in network performance. For example, if a user connected to the server via a slow network link requests a document or information,
Users may experience significant delays. As the web transfers information to users over the network, the level of traffic in the network is further increased. The increase in traffic reduces the bandwidth available for other users' requests, causing further increase in potential delay.

To reduce client access latency, the prior art stores or stores a copy of the popular information document closer to the user at the network node, thereby making the access latency more acceptable. It became so. Caching can be implemented at various points in the network. For example, a large university or company has its own local cache from which all users subscribing to the network can retrieve documents. local·
The cache can be implemented in the form of a special server.

A special server, called a caching proxy, can act on behalf of a client as a proxy for looking for a potential cached copy of the requested information. Because the caching proxy is only concerned with misses left from the failed client cache, the caching proxy typically operates as a secondary or higher level cache. The client cache is embedded in various web browsers. The client cache can either store only the document access during the current call (a non-persistent cache like a mosaic) or store the document across all calls.

Previous work on caching web documents was primarily in the area of on-demand caching. On-demand caching means that the caching decision is made after the object has been requested. For example, Abrams et al., "Caching Proxy: Limits and Potentials (Caching
Proxies: Limitations and Potentials ”(Report of the Fourth International World Wide Web Conference 1996) describes various on-demand caching policies that have been considered for the Web.

[0017]

In an Internet Service Provider (ISP) network or an enterprise network, different regional or departmental proxies in the proxy hierarchy may observe different referral behaviors and / or different network traffic and traffic. Face delay. The existence of a proxy hierarchy further complicates the prefetch decision as to whether all intermediate proxies, clients, or a subset thereof, should perform prefetch. Accordingly, there is a need for a prefetch method and system that can effectively coordinate prefetch operations with respect to a proxy server hierarchy and associated clients.

[0018]

[Means for Solving the Problems]

Accordingly, the present invention provides for pre-fetching selection information from a content server so that one or more clients associated with the content server can retrieve the selection information via a communication network and one or more proxy servers. Requesting information from at least one of a proxy server; and retrieving information from the client based on a past search or access to the requested information by any of the clients. Establishing pre-fetch hint information for requesting; annotating the request information with the pre-fetch hint information; based on the pre-fetch hint information and the request information; Retrieving.

Generally, the request information includes a request object, and the selection information includes an Internet
Contains pre-fetch candidate objects such as data. In this case, the request object can reference the selected object via a hypertext transfer protocol (HTTP) link associated with the Internet data, and the proxy server sends the request object stored in the request object's object header. Identify the relationship of the selected object to the request object through the object's Universal Resource Locator or URL's HTTP protocol.

Establishing prefetch hint information comprises: tracking a data reference pattern associated with the client; and organizing the data reference pattern into a group of object identifiers by a relationship between the request object and the prefetch candidate object. Determining which of the set of identifiers has a sufficient relationship between the pre-fetched candidate object and the request object to ensure that the pre-fetched candidate object is pre-fetched as selection information. Determining. Thus, the data reference pattern of the client is advantageously based on a previous search or previous access to the requested information, and the data reference pattern is determined by various ones of the clients associated with one of the proxy servers. Selected to represent general trends in past searches or accesses. This can then be used to define prefetch hint information.

The tracking and organizing steps comprise establishing a database for storing reference data patterns in the proxy server, and in each of said databases, an object identifier that distinguishes a request object from another request object. Placing an object count indicating the number of times the requested object has been requested as observed by one of the servers; and one or more pre-fetch candidate objects associated with each requested object Preferably comprises associating the access list contained in each with the database.

Such a database can be used to define the relationship between a request object and a pre-fetch candidate object and track access statistics of the request object and the pre-fetch candidate object. The access statistics include a parent object access frequency of the request object and a child object access frequency of the prefetch candidate object. Parent object access frequency refers to the number of times a particular one of the requested objects is requested by at least one of the clients. The child object access frequency represents the number of times one or more of the pre-fetch candidate objects are actually requested before or after the particular one is accessed. This allows child objects
Based on the ratio of child object access frequency to parent object access frequency associated with access frequency, it is possible to select a pre-fetch list of pre-fetch candidate objects containing only pre-fetch candidate objects with a ratio exceeding the threshold probability become.

In general, at one of the proxy servers or one of the clients,
Only a subset of the pre-fetch candidate objects are actually pre-fetched and received. The selection of this subset of prefetched candidate objects as selection information is based on factors such as the free storage capacity at one of the proxy servers and / or the number of accesses at one of the proxy servers. Thus, pre-fetch candidate objects are pre-fetched according to priorities associated with some corresponding pre-fetch values.

In a preferred embodiment, the step of annotating the request information comprises:
Placing the prefetch hint information on a prefetch label having a platform for content selection (PICS) format or a label protocol compatible with the PICS format. In a particular implementation of this (PICS) format, an object identifier category and a referrer identifier category are defined, and the method further comprises: referencing the referrer identifier category to identify the actually referenced prefetch candidate object. Including returning to the server. The organization of pre-fetch hint information by a platform protocol for Internet content selection or a PICS compatible protocol utilizes a format that is already supported by many browsers.

Generally, there is a hierarchy of the proxy server. It is preferable to define a transmitting node and a receiving node in this hierarchy and change the prefetch hint information from an initial state to a final state, where the initial state is determined by the transmitting node and the final state is determined by the receiving node. It is then the prefetching hint information of the final state that forms the basis of the final prefetching decision (assuming that the prefetching of the selection information is initiated by the receiving node).

In other words, the prefetch hint information can be interpreted and updated as it is transmitted from the proxy server to the client via the hierarchy. The proxy server preferably updates the prefetch hint information whether there is a local reference to the request information and whether the count of the local level server exceeds the minimum threshold count (i.e., a reliable statistic Is available locally)
. The prefetch hint information may also be updated based on the caching state in the proxy hierarchy (eg, indicating where in the hierarchy a particular prefetch candidate object may be cached).

[0027] The present invention further provides one or more proxy servers associated with the network.
A data processing system for prefetching selection information from a content server over a network, the server including a server and a client for communicating with the content server and one or more of the proxy servers over a network. Means for requesting information from at least one of the proxy servers; and means for establishing prefetch hint information for the client based on past retrieval or access of the requested information by any of the clients. A data processing system comprising: means for annotating the request information with the prefetching hint information as an annotation; and means for preliminarily extracting selection information from any of the servers based on the prefetching hint information and the request information. I will provide a.

In a preferred embodiment, the request information includes a request object and the selection information includes a prefetch candidate object, wherein each of the proxy servers and at least one of the clients includes a main cache, a prefetch cache, Wherein the prefetched selection information is stored in a prefetch cache.

The prefetching means prefetches the object based on a priority configuration in which the object having the highest prefetch value is stored first with respect to the free storage capacity of the prefetch cache, and deletes the prefetched object. Preferably, the prefetch values are stored in a prefetch cache, based on individual or collective reference probabilities to the prefetch object by one or more clients. In particular, the prefetch value is based on an estimated access probability that the prefetch candidate object on the prefetch hint information list will be accessed by a particular client, and a required storage capacity for storing the selected one of the prefetch candidate objects. It can be based on the access cost of the object.

In a preferred embodiment, the pre-fetching means comprises a pre-fetching handler including a calculating means for calculating the pre-fetching value and an evaluating means for evaluating the pre-fetching value calculated by the calculating means. Wherein the calculating means calculates a prefetched value of each prefetched object identifier on the prefetching hint information list, the evaluating means evaluates a prefetched value of the plurality of candidate prefetched objects, and prefetches. Prioritize caching of candidate prefetched objects associated with the highest prefetched value on the hint information list. The prefetch means may include a first prefetch handler in at least one of the proxy servers and a second prefetch handler in at least one of the clients (ie, at different / multiple locations in a hierarchy). Can be taken out in advance)
.

A preferred embodiment also includes a client object request handler stored in one of the clients' memories. The client object request handler searches the main cache and prefetch cache for the request object before requesting the request object from the proxy server or the network. The client object request handler determines whether the object is in the prefetch cache by generating a prefetch reference message to the server or server indicating that the prefetch object was actually accessed by the client. Or notify the proxy server. Similarly, at least one of the proxy servers includes a prefetch reference message handler for forwarding the prefetch reference message after invoking the cache manager routine if the request object is in the prefetch cache. It is preferable to include This ensures that the server can maintain an accurate record of access statistics, as it is obtained from the prefetch cache of the client or proxy server, even if it does not communicate directly with the child object.

The present invention further provides a content server or proxy server for a data processing system in which the request information includes a request object and the selection information includes a prefetch candidate object, wherein the parent object identifier and the related child object identifier. Statistic table that stores a first count of parent objects representing the frequency with which parent objects are referenced by a client, and that related child objects are referenced after a selected one of the parent objects is referenced. There is provided a server comprising: the statistical table including a second count of a child object representing a frequency; and a generating unit configured to generate a prefetching hint information list based on information in the statistical table.

In one embodiment, the generating means comprises a first count, a second count,
Accepting input of a child object identifier and a parent object identifier, the generating means includes evaluating means for evaluating an access relationship between the parent object and their associated child objects, the evaluating means including: If the child object is immediately referenced at a suitable frequency immediately before the client's reference, the child object identifier is placed in the prefetch hint information list. This appropriate frequency may be based on the ratio of child access to corresponding parent access.
Generally, it calculates the ratio of the second count to the first count for each of the parent object identifiers, compares this ratio to a hint threshold, and if the ratio exceeds the hint threshold, the child object and the associated parent object And grouping or associating with.

[0034] Preferably, the server further comprises statistics updating means for updating the statistics table based on the actually requested child object, said statistics updating means first comprising: The identity searching and referencing or referrer objects are referenced before the actual requested child object, which in turn causes the corresponding second count to be updated in the statistics table. The statistics table must also be updated by receiving a prefetch reference message from the prefetch reference message handler.

If the server is a proxy server, it preferably further comprises a prefetch update means for receiving and updating the prefetch hint information list. The prefetch update means compares the first count with the first local threshold count and calls the generating means for each child object identifier whose first count exceeds the first local threshold count. If the first count does not exceed the first local threshold count,
The prefetching update means maintains the original prefetching hint information list. In a more sophisticated method, the prefetch update means compares the first count with the first and second local threshold counts, and (i) if the first count does not exceed the first local threshold count, the original prefetch hint. Maintaining the information list, (ii) if the first count exceeds the first local threshold count but does not exceed the second local threshold count, modify the original prefetch hint information list, and (iii) the first count If both exceed the first and second local threshold counts, exchange the original pre-fetch hint information list.

The server is further a proxy object handler for returning the request object to a lower order proxy server or client that has requested the request object, and returns the request object if the request object is not a prefetch object. And said proxy updating the statistics table accordingly.
Preferably, it includes an object handler and a prefetch cache manager for managing the prefetch cache. Therefore, the proxy object handler returns the requested object from the prefetch cache to the requesting client or proxy server and allows the object to be deleted from the prefetch cache if the object was requested in accordance with the prefetch request. To mark. If the request object is not in the main cache or the prefetch cache, the proxy object handler sends the object request to another server. If the request object is not associated with a prefetch request, and if the request object is in the prefetch cache or main cache, the proxy object handler updates the statistics table accordingly.

[0037] In another aspect, the present invention provides a method for selecting a selected object from a content server to facilitate the availability of the selected object to a client associated with one or more proxy servers via a communication network. A) requesting a request object from at least one of said proxy servers; and b) retrieving a corresponding client based on a past search or access of the request object by the client. Identifying a data reference pattern; c) creating prefetch hint information from the data reference pattern into a prefetch hint information list; d) annotating the request object with the prefetch hint information;
e) Based on the pre-fetch hint information and the currently requested object,
Pre-retrieving a selected object from at least one of said servers.

Thus, the prefetched selection information comprises a prefetch candidate object, and the step of establishing the prefetch hint information further comprises establishing the prefetch hint information at a sending node in the hierarchy of the proxy server. Pre-fetch candidate objects are passed through the proxy server hierarchy, and the selection information for pre-fetching at the receiving node in the hierarchy is based on the pre-fetch hint information.

In general, there is a proxy server hierarchy associated with the network, which includes lower-level servers that are more closely associated with clients than other proxy servers. In one preferred embodiment, the prefetch hint information is updated via the proxy server hierarchy if there is a local reference to the request object and if the local level server count exceeds the minimum threshold count. . Alternatively (or in addition), the prefetch hint information can be updated by analyzing the caching state higher in the hierarchy. The top of the hierarchy is associated with the network closest.

In another aspect, the invention is a processing system for enhancing the flow of data from a server to a client along a network, comprising: a proxy server associated with the network; and a content server associated with the network. A server, a client configured to communicate with one or more of the proxy servers, and a statistics table stored in at least one of the proxy servers.
A parent object identifier and an associated child object identifier, including a first count of parent objects representing a frequency with which the parent object is referenced by the client, the corresponding child object after a selected one of the parent objects is referenced Said statistic table including a second count of child objects representing the frequency with which is referenced, and prefetching hint information by accepting inputs of a first count and a corresponding second count to derive a prefetching hint information list. Generating means for generating a list, for each of the parent object identifiers, calculating means for calculating a ratio of a second count to a first count; and comparing means for comparing the ratio with a hint threshold. And if the ratio exceeds the hint threshold Comparing means for grouping or associating objects and related parent objects on a pre-fetch hint information list; and pre-fetch means for pre-fetching child objects based on the pre-fetch hint information and the parent object actually requested by the client. And a generating system comprising:

In a preferred embodiment, each of the proxy servers has a first processor, a first memory, and first storage means for storing data, wherein the first memory has a first main cache and a first memory. Has at least a logical separation from the prefetch cache,
The content server has a second processor, a second memory, and second storage means for storing data, and at least one of the clients has a third processor, a third memory, and a third memory for storing data. Storage means, the third memory having at least a logical separation between the second main cache and the second prefetch cache, and the prefetch processing means having child storage in the first prefetch cache or the second prefetch cache. Stores objects.

In a preferred embodiment, the pre-fetching means includes calculating means for calculating a pre-fetched value, and evaluating means for evaluating the pre-fetched value calculated by the calculating means, wherein the calculating means includes: The prefetching value of each child object identifier on the prefetching hint information list is calculated, and the caching of the child object having the highest prefetching value on the prefetching hint information list is prioritized. To manage free space.

The system further includes prefetch update means for updating the prefetch hint information list, the prefetch update means comparing the first count with the first local threshold count and the second local threshold count, and updating the first local threshold count. The threshold count is based on statistics from one server, the second local threshold count is based on statistics from another server, and the prefetch and update means is configured to determine that the first count is a first local threshold count and a second local threshold count. Preferably, the generating means is called for each of the child object identifiers exceeding both of them, and the prefetch update means maintains the original prefetch hint information list if the first count does not exceed the first local threshold count. .

[0044] The system includes a statistics update unit for updating a statistics table, and a proxy object handler stored in one of the proxy servers, which calls a prefetch hint information list update routine and The proxy object handler returns a request object to a lower-level proxy server when the proxy server requests the request object, and returns the request object when the request object is not a prefetched object and calls the statistics update means. A prefetch cache manager for managing an associated prefetch cache of one of said proxy servers, prior to being called by a proxy object handler if a subordinate server did not request a request object. An article prefetch cache manager and a proxy object request handler stored on one of the proxy servers, wherein the object is in a prefetch cache and when the object is requested according to a prefetch request. Returns the object to the requesting client of the clients and marks the object as removable from the prefetch cache, and if the requested object is not in the main cache or the prefetch cache, sends the object request to another server. If the request object is not associated with a prefetch request, and if the request object is in the prefetch cache or the main cache, a proxy object request handler that communicates with the statistics updater is used. It is preferable to include in.

The system further includes a statistics update unit for updating a statistics table, and a prefetch reference message handler stored in one of the proxy servers, wherein the call unit calls the statistics update unit, and the request object is And a prefetch reference message handler that forwards the prefetch reference message after invoking the cache manager routine if it is in the prefetch cache.

The present invention is further a processing system for enhancing the flow of data from a server to a client along a network, comprising: a proxy server associated with the network; a content server associated with the network; A client configured to communicate with one or more of the proxy servers, and a statistics table stored on at least one of a content server and the proxy server, comprising a parent object identifier and an associated child object. An identifier is stored and includes a first count of the parent object representing a frequency with which the parent object is referenced by the client, and corresponding child objects are brought into close proximity in time after a selected one of the parent objects is referenced. Shows the frequency of reference Generating means for receiving input of a first count, a second count, a child object identifier, and a parent object identifier from the statistical table, the statistical table including a second count of child objects; Evaluation means for evaluating an access relationship between the child object and the parent object before and after the client refers to the corresponding parent object. The generation means for arranging the child object identifier in the pre-fetch hint information list; and a pre-fetch means for pre-fetching the child object based on the pre-fetch hint information and the parent object actually requested by the client. And a processing system including:

In a preferred embodiment, the pre-fetching means further comprises an assigning means for assigning a storage priority to the child objects identified on the pre-fetching hint information list;
Selective storage means for storing a selected one of the child objects according to the storage priority. The allocating means is based on an estimated access probability that the child object on the pre-fetch hint information list is accessed by a specific client and an access cost of the child object based on a required storage capacity for storing a selected object among the child objects. It is preferable to assign storage priorities.

In this way, information can be pre-fetched at the proxy layer and the object access time via the client's network (ie, the Internet) can be reduced. The selection information is pre-fetched from the content server so that clients associated with the content server in the network can easily access the selection information. The client is preferably associated with a proxy server on the network. The client requests information from at least one of the proxy servers. The content server generates pre-fetch hint information of the client based on the past data access pattern of the information requested by the client. The content server annotates the request information with prefetching hint information. One or more proxy servers can dynamically update the prefetch hint information based on the dynamic usage statistics. As the object passes through the proxy hierarchy, the prefetch hint information is updated to reflect the prefetch operations performed, the state of caching above the proxy hierarchy, and other local considerations such as local reference patterns. You. The selection information can be pre-fetched from the content server or any proxy server based on the pre-fetch hint information and the requested information. A prefetch value can be assigned to the prefetch hint information based on the reference access probability and storage considerations. The selection information with the highest prefetch value in the prefetch hint information list is stored in the cache first, and potentially until the prefetch cache is full.

The preferred embodiments of the present invention will now be described in detail, by way of example only, with reference to the accompanying drawings.

[0050]

BEST MODE FOR CARRYING OUT THE INVENTION

FIG. 1 shows a client 22, a hierarchy of proxy servers 24, a network 25.
, And a data processing system including a content server 20.

The client can request information in the form of a parent object, a child object, a request object, a selection object, a page, request information, and selection information. Parent objects are associated with child objects so that one or more related child objects can be requested after the parent object has been processed historically. Requested objects are similar to parent objects, while selected objects are similar to child objects. A request object is an object that is actually requested by one or more clients. on the other hand,
A selection object is an object that is grouped with or associated with the request object so that the selection object can be requested in close proximity after the request object. The request information is a request object,
Contains the parent object and other forms of data required by the client. Selection information includes child objects, selection objects, and other forms of data that can be retrieved in advance.

The client 22 can request information from the content server 20 or the proxy server 24. However, proxy server 24 typically operates as an intermediary for client 22 to network 25 and content server 20. The proxy server 24 may provide the content server 2 if the proxy server hierarchy contains the requested information or has the ability to complete the requested task.
Advantageously, the client information request can be satisfied without contacting 0. The data processing system of FIG. 1 can be applied to the Internet.
For example, content server 20 may include a web site, while network 25 may include the Internet.

FIG. 1 illustrates logical connections 26 rather than physical connections between elements of a data processing system.
Is shown. Logical connection 26 represents the flow of object and request information within the data processing system. Logical connection 26 generally does not represent a physical connection between proxy servers 24. The logical connections 26 may change due to the amount of work and / or physical events such as node or communication link failures. Different types of objects may follow different logical transmission paths until reaching the client 22.

While in practice practically any number of clients 22 and proxy servers 24 can be used, the example of FIG. 1 comprises four clients 22 and four hierarchical levels of proxy servers 24. . In fact, any number of levels of proxy server hierarchy is possible, and clients 22 can logically connect to any level of the proxy server hierarchy. For example, proxy server 2
The four tiers can be a single proxy server. The present invention may be implemented with only a single client 22 and a single proxy server 24.

As shown, the top-level proxy server is designated as level 0 proxy server 30. "Level 0" is any symbol that simply defines the state of one proxy server relative to another proxy server in the example of FIG.
The top-level proxy server is logically connected to the network 25. The upstream proxy server 21 is closer and logically associated with the network 25 than the lower proxy server 23. (Thus, in this terminology where a level 0 proxy server is considered the top of the proxy hierarchy, level 0 is more "
High ", level 1 is" higher "than level 2, and so on. )

The first proxy server 35 and the second proxy server 37 at the level 1 are logically coupled to the proxy server 30 at the level 0. The first proxy server 35 of level 1 is connected to the first proxy server 40 and the second proxy server 43 of level 2. Level 2 first proxy server 40 is logically coupled to level 3 first proxy server 50 and second proxy server 55.

The client 22 includes a first client 600, a second client 603, a third client 700, and a fourth client 703. The first client 600 and the second client 603 are the first level 3 proxy server 5
Logically connected to 0. Third client 700 and fourth client 70
3 is logically linked to the second proxy server 55 at level 3.

Now, the data processing system will be described from the viewpoint of the second client 603.
The second client 603 is connected to the next higher level proxy. As shown, the next higher level proxy is a level 3 first proxy 50.
Next, the second client 603 is connected to the first proxy server 40 at level 2, the first proxy server 35 at level 1, and the proxy server 30 at level 0, respectively. The second client 603 can access various content servers 20 via the network 25. Second client 60
The third general communication path is via the next higher level proxy (ie, the first level 3 proxy server 50), while the second client 603 or lower level proxy is the other higher level proxy server. 24 or network 25
Can communicate directly. The second client 603 can also have its own client proxy to manage caching.

From the perspective of the second client 603, a particular proxy server is not part of its proxy hierarchy. For example, level three second proxy server 55, level two second proxy server 43, and level one second proxy server 37 are not part of the second client proxy server hierarchy.

[0060] The client 22 may include, but is not limited to, a personal computer, a workstation, a set top box, and the like. Network 25 may include, but is not limited to, the Internet, the World Wide Web, an intranet, a local area network (LAN), and the like.

The proxy server 24 and the content server 20 can include various commercially available computers. For example, the proxy server 24 uses IBM
Internet Connection Server sold by Corporation (
ICS). Similarly, content server 20 may be implemented as a Lotus Go Web server, a Lotus Domino server, or the like (also commercially available from IBM Corporation). The proxy server 24 or the content server 20 uses S / 390 SYSPLEX,
It can be run on any compute node including, but not limited to, a product such as SP2, or an RS / 6000 workstation (also available from IBM Corporation). In summary, proxy server 24 or content server 20 includes any general-purpose computer and associated software that can provide access to world wide web pages, remote file transfer, e-mail, and transaction support to client 22. be able to.

The data processing system of FIG. 1 can maintain data transfer efficiency using the attributes of the conventional proxy layer. For example, if the request object is not available locally at the lower proxy server, the request for the unavailable object is made to the next higher proxy. If the upstream proxy has previously stored the object in its cache memory, the upstream proxy hands down the requested object to the client via the downstream proxy server. Otherwise, this upstream proxy server will try to get the object from the next higher upstream proxy. Once the request object is obtained, it is passed down to the lower level proxy or client that requested the object.

FIG. 2 shows a more detailed example of the data processing system shown in FIG. The data processing system includes a proxy server 24, a content server 20, a network 25, and a client 22. The data processing system includes a prefetch unit 250, a statistics table 261, a statistics update unit 252, and a main cache 29.
7, and a prefetch cache 298.

The prefetch unit 250 preferably includes a first prefetch handler 248 in the proxy server 24 and a second prefetch handler 241 in the client 22. The statistics table 261 includes a first statistics table 293 associated with the proxy server 24 and a second statistics table 267 associated with the content server 20. The statistics update unit 252 includes a first statistics table 293
Update statistic unit 264 for updating, and second statistic table 267
And a second update statistics unit 269 for updating. The main cache 297 is
First main cache 294 in each proxy server 24 and each client 22
Within the second main cache 246. The prefetch cache 298 includes a first prefetch cache 296 in the proxy server 24 and a second prefetch cache 249 in the client 22.

The proxy server 24 is a computing node that can satisfy a request via the network 25. The proxy server 24 includes a first processor 280, a first memory 290, and a first storage device 281. The first memory 290 can include, for example, a random access memory (RAM). The first storage device 281 may include, for example, a direct access storage device (DASD). First memory 2
90 is a proxy server logic 2 implemented as computer executable code
95 is stored. Proxy server logic 295 is loaded from first storage 281 to first memory 290 for execution by first processor 280.
Details of the operation of the proxy server logic 295 are shown in FIG.

The proxy server logic 295 controls the proxy object request handler 283
, Proxy object handler 284, first prefetch handler 248,
Pre-fetch reference message handler 285, pre-fetch update means 291, first
Update statistics routine 264, first statistics table 293, first main cache 294,
And a first prefetch cache 296.

FIG. 12 shows details of the operation of the proxy object request handler 283. Details of the operation of the proxy object handler 284 are shown in FIG. The operation of the prefetch reference message handler 285 is shown in FIG. Statistics update unit (
That is, details of the operation of the first update statistical routine) are shown in FIG.

The operation of the prefetch unit 250 (ie, the first prefetch handler 248) is shown in detail in FIG. The operation of the pre-fetching update unit 291 or the pre-fetching hint information updating routine is shown in FIG. The proxy server logic 295 determines the first
Main cache 294 and first prefetch cache 296 are maintained.

The content server 20 is a computing node that can satisfy the request via the network 25. The content server 20 includes a second processor 260 and a second
A memory 263 and a second storage device 265 are included. The second memory 263 can include, for example, a random access memory (RAM). Second storage device 26
5 may include, for example, a direct access storage device (DASD). Second memory 263 stores content server logic 268 or content server software.

The operation of the content server logic 268 is described in more detail below in connection with FIG. Content server logic 268 is implemented as computer-executable code that is loaded from second storage 265 into second memory 263 for execution by second processor 260. The content server logic 268 includes a second statistics table 267, a second update statistics unit 269 for updating the second statistics table 267, and a generation unit 2 for generating prefetch hint information.
66. The second statistics table 267 stores data regarding the information requested by the client, and potential prefetch information associated with the requested information. The statistics table 261 contributes to generation of prefetching hint information based on an actual reference operation of the client. Details of the statistics table 261 are shown in FIG.

The operation of the second update statistics unit 269 is shown in more detail in connection with FIG. The generation unit 266 generates pre-fetch hint information from the information in the second statistical table 267 and the request information. The generation unit 266 can be implemented as a pre-fetch hint information (PHI) generation routine as shown in FIG.

The client 22 includes a third processor 240, a third memory 245, and a third
Preferably, the storage device 242 is included. The third memory 245 includes, for example, a random
An access memory can be included. The third storage device 242 may include, for example, a direct access storage device (DASD). Third memory 245 stores client logic 244 or client software. The operation of the client logic 244 will be described later in more detail in connection with FIG. Client logic 24
4 is implemented as computer-executable code, which comprises a third processor 24
0 to be loaded from the third storage 242 into the third memory 245 for execution.

The client logic 244 is a client object request handler 247
, A second prefetch cache 241, a second main cache 246, and a second prefetch cache 249. Client object request handler 247
Obtains the requested object for the client from the second main cache 246, the second prefetch cache 249, the proxy server 24, or the content server 20. The details of the operation of the client object request handler 247 are shown in FIG.

The client logic 244 maintains a second main cache 246 and a second prefetch cache 249. The second primary cache 246 is used to store frequently referenced objects, while the second prefetch cache 249 is used to store objects before they are actually referenced. A pre-fetched object is an object that has not been previously or previously referenced in relation to a reference object. In the preferred embodiment, the primary cache 29
7 is managed by a conventional LRU (least frequently used) algorithm. Those skilled in the art will also appreciate that any other replacement strategy can be applied to maintain the main cache 297.

The main cache 297 and the prefetch cache 298 can be logically separated. Logically separating prefetched objects from the actual referenced objects in main cache 297 prevents the prefetched objects from taking too much storage space. Various alternatives are available to achieve this goal of a logically separate prefetch cache and main cache. For example, a single cache may impose an upper limit on the number of prefetched objects and / or impose an upper limit on the space occupied by prefetched objects. Main cache 297 and prefetch cache 298 are intended to support different goals. The main cache 297 is to store popular objects that have the potential to be repeatedly referenced. The prefetch cache 298 is only used to support a particular browsing session by prefetching the object that is most likely to be referenced next in that session. In a preferred embodiment, when a lower node of the hierarchy issues a prefetch request for an object in the higher node's prefetch cache, the object is dropped down to the lower node and the higher node Are marked as deletable from the prefetch cache 298. When a prefetched object is actually referenced, it can be moved to the main cache 297 according to the management policy of the main cache 297. In a preferred embodiment using the LRU replacement policy, once fetched prefetched objects are moved to the main cache.

Alternative criteria can be devised to allow the movement of objects from the prefetch cache 298 to the main cache 297. Abrams et al., "Caching Proxies: Limitation and Potential"
Potentials) (Report of the 4th International World Wide Web Conference 1996) discusses several cache replacement policies that take into account the size of objects in the decision-making process. Other cache replacement strategies are known to those skilled in the art.

FIG. 3 is a flowchart illustrating a method for dynamically pre-fetching data for a client from a content server 20 or proxy server hierarchy. Initially, at block 100, a client requests information from a proxy server or content server 20. Second, at block 102, patterns of historical data access or data reference at the content server 20 and the proxy server are identified. In practice, the order of completing blocks 100 and 102 can be reversed.

Third, at block 104, the content server 20 generates prefetch hint information (PHI) for related access to the request information based on the data reference pattern identified at block 102. Fourth, at block 105, pre-fetch hint information may be annotated with the request information. For example, the pre-fetch hint information can be transmitted as meta data added to the request information (ie, the object). Regarding the step of establishing or generating prefetch hint information, the content server 20 tracks usage / reference patterns from multiple concurrent users or clients. In effect, the content server 20 determines which object group is associated with the request information or request object. The content server 20 generates prefetch hint information (PHI) and provides it to the request object. At or shortly before the prefetch step, the proxy server can interpret the PHI in light of local considerations to determine which selected objects to prefetch.

Fifth, at block 106, as the request information passes through the proxy server hierarchy, the prefetch hint information is updated. The update at block 106 reflects the prefetch operation to be performed and the caching state at higher levels, as well as other local considerations, such as local reference patterns. The proxy server can modify the PHI to reflect its pre-fetching decisions and various local considerations before passing the object to the next (lower) level proxy or client station. When an object passes through the proxy hierarchy, the sending node sets the initial PHI while the receiving node sets the initial PHI.
The final PHI can be derived from the HI. The final PHI determines which selected objects to prefetch based on the requested objects. Pre-fetch operations and decisions are triggered at the receiving end, not at the transmitting end.

Finally, at block 108, the selection information is pre-fetched based on the pre-fetch hint information and the request information. The proxy server or client may use the PHI information in the received object based on the PHI information, locally maintained reference information (if available), and other local considerations such as free storage space and access time. Determines whether to prefetch a subset of the objects specified in.

The generation of prefetch hint information in block 104 preferably uses the PICS protocol or PICS compatible format (see below). The PHI generated at block 104 is used to store each cached object so that when the cached object is requested, the PHI information for that object can be looked up and appropriate prefetch operations can be triggered. Is stored as part of the meta data together with the PHI information can be provided by the PICS protocol. The PICS protocol provides information about the severity of each PHI object and its characteristics, such as, for example, size, as well as higher-level caching state.

Updating of the pre-fetch hint information can be performed through the proxy server hierarchy. The proxy server can collect local information and distribute or distribute pre-fetch decisions down the proxy hierarchy. Further, the information of which intermediate proxies have pre-fetched the objects may have implicit meaning to the pre-fetching decisions of lower-level proxies and client stations in order to optimize use for storage location and bandwidth.

The proxy servers in the proxy server hierarchy work together to provide receiver-oriented pre-fetch decisions. Prefetching decisions are made jointly by each proxy and client. The pre-fetch hint information in the form of meta information becomes an annotation (ie, is appended) to the request object. Prefetch hint information provides information about potential prefetch candidates. The intermediate proxy server may change the meta-information or pre-fetch hint information based on local state, so that pre-fetch can be triggered when referencing the object later.
Stores meta information with the object when it is stored in the cache.

FIG. 4 shows a data format of the statistics table 261. Statistics table 2
61 generally refers to either the first statistics table 293 or the second statistics table 267. One or more proxy servers 24 maintain a first statistics table 293. The content server 20 maintains a second statistical table 267. The format of the first statistic table 293 and the second statistic table 267 is the same as that of the statistic table 2 in various proxy servers in the proxy server hierarchy.
Preferably, they are substantially similar or identical to facilitate updating of 61.
The statistics table 261 serves to provide prefetch hint information (PHI) based on actual viewer browsing behavior from the perspective of the proxy server or content server 20 with which the statistics table 261 is associated. The statistics table 261 can observe all reachable client references below it in the proxy server hierarchy.

For example, in FIG. 1, the first statistics table 293 in the first proxy server 50 at level 1 tracks the aggregate data access operation of the first client 600 and the second client 603. On the other hand, the first statistical table 293 in the second proxy server 55 at the level 3 indicates that the third client 700 and the fourth client 70
3 aggregate data access operations. Level 2 first proxy server 4
The first statistical table in 0 is the first client 600, the second client 603
, The third client 700, and the fourth client 703. Lower-level proxies such as the first proxy server 50 at level 3
00 may be able to better track the interest, but the lower level proxy is
Significant advance retrieval hint information (PHI) from clients of sufficient sample size
May not be able to get enough reference statistics to provide

In the preferred embodiment, when the object O is requested from the content server 20, the server 20 provides the initial prefetch hint information PHI based on the information in the second statistical table 267. With enough local references to object O, when object O is passed down the proxy hierarchy, proxy server 24 will prefetch the hint information (PHI) in the object header along the path to client 22. ) Can be changed to adapt the PHI to the local information in the first statistical table 293. The first statistical table 293 includes a lower-level proxy
The second statistics table 267 can be replaced based on factors such as the well-indexed interest or preference of a particular client by the server.

In the preferred embodiment, the content server 20 maintains a second statistics table 267, which is a database of client 22 data access frequencies for request information. The database can be extended to include the requesting proxy server 24 data access frequency. Similarly, one or more proxy servers 24 maintain a first statistics table 293. For example, the statistics table 261 includes a first array 3 having a first data structure 304 and a second data structure 306.
It has 83. As shown, the first data structure 304 and the second data structure 30
6 is a record. Each record of the statistics table 261 includes several fields including a parent object identifier (object ID) 301, a first count (T count) 302, and an access list data pointer (A list) 303.

The parent object identifier field 301 contains an object identifier. For example, parent object identifier 301 includes the URL of an object for the Internet or related application. The first count field 302 contains the number of times the parent object associated with the parent object identifier has been referenced, as observed by the server. That is, the first count field 302 contains the actual request frequency of the request object. The first count 302 indicates a parent object access frequency. The access list data pointer 303 contains a data pointer to the access list 370.

The statistics table 261 has an access list 370, which includes, for example, a first access record 340, a second access record 350, a third access record 360, and a fourth access record 380. As shown, the first
The parent object identifier associated with the data structure 304 has an access list 370 consisting of a linked list of a first access record 340, a second access record 350, and a third access record 360. The next record data pointer field 313 contains a pointer to the next access record on the access list 370 associated with the specific parent identifier 301. Second
Parent object identifier 301 associated with data structure 306 only includes fourth access record 380.

Each record or data structure on the access list is composed of several fields including a child object identifier 310, a second count (R count) 312, and a next record data pointer (NXT) 313. . The data structure is
It can be in the form of a general data structure, record, or array.
As shown in FIG. 4, the data structure is the records that form the linked list in the access list 370.

In an alternative embodiment, the statistics table data format may include a primary array associated with a plurality of alternative secondary arrays in a tree structure. The primary array contains information related to the parent object, and the secondary array contains information about child objects of the corresponding parent object.

[0092] The child object identifier field 310 contains the object identifier (URL) of the object. The parent object identifier 301 is 1 in the access list.
Associated with one or more child object identifiers 310. The child object identifier 310 identifies a potential pre-fetch candidate if the child's parent object is actually referenced. The second count 312 includes the number of times a child object is referenced through its parent object. In other words, the second count is the frequency with which the child object is referred to in a temporally close manner after the prior reference of the associated parent object. The second count 312 indicates a child object access frequency.

The child object identifier 310 on the access list 370 can be obtained dynamically based on observation of a client's actual browsing operation. For example, HTM
In L, a web page can have HTTP links to other web pages. The web page pointed to by the HTTP link is a potential child object of the (parent) web page that contains the HTTP link. When a parent object is displayed and a reference (HTTP) link included in the parent object is clicked, a corresponding child object is referenced. Because a reference link to a child object may be included in multiple other objects, a child object can have more than one parent. In the HTTP protocol, a request for a child object includes a parent URL as reference side information in a header thereof. Thus, the server can dynamically associate subsequent child references to its parent (by referrer information in the child's header). If the client frequently clicks or selects one child object, that child object can be a pre-fetch candidate or selection information.

In addition to dynamically tracking viewer browsing activity, statistical analysis or mining of web logs can be performed periodically to derive information about groups of objects for the statistics table 261. . The mining algorithm identifies a group of objects that are referenced together to obtain potential prefetch candidates for the PHI. One way to mine web logs is to use SpeedTracer: Web Usage Mining and Analyze by Kun-lung Wu et al.
sis Tool ", IBM Research Report 20895, May 1997 (IBM
Systems Journal v37, nl, p 89-105, 1998). The SpeedTracker algorithm provides information about frequent page groups and frequent search paths that are referenced together in a user session. The SpeedTracker algorithm is an advanced type of mining that can find related references that are not directly referenced together. It can also make good predictions for the next set of reference objects based on the search path to any object.

Using another kind of data mining algorithm, the statistics table 26
One data access pattern can be identified. For example, other data mining algorithms include classification and clustering, which are
It can be applied to log analysis to obtain reference patterns. In addition, a PHI can also be provided by the content server based on its knowledge of object relationships. For example, the PHI may include patches and upgrades for required software downloads, Java classes of required JAVA applets, and objects (such as GIFs) contained within the request object,
It is not limited to these.

In a preferred embodiment, when collecting statistics, the actual reference of an object by a client station is distinguished from a pre-fetch request for the object. The first count 302 and the second count 310 in the statistics table 261 reflect actual references rather than prefetching. That is, as a result of the pre-fetching operation, the first statistics table 293 or the second statistics table 267 is not updated, but is updated only by the actual object reference.

The pre-fetch hint information or pre-fetch hint information list may use various data structures to annotate the request information or request object. For example, the pre-fetch hint information can be exchanged between the requester and the server node using the added metadata. In an HTTP implementation, the information exchange uses existing web protocols to transfer meta-
Can be included as data. PICS ("Platform for Internet Content Selection") specifies how to transmit meta-information about electronic content. PICS uses (a
It can be applied to annotating request information, (b) pre-fetch selection information, pre-fetch candidates, or child objects, and (c) confirming pre-fetch operations after the information is pre-fetched.

PICS is known to those skilled in the relevant art as a protocol recommendation of the Web Consortium. PICS was initially used to send rating labels based on value, such as "how much nudity exposure is associated with this content," but the format and meaning of the meta-information is completely universal. is there. PI
In CS, meta-information about electronic content is grouped according to the purpose of use with the "rating service" or the creator of the information, and within such one group any number of categories or sizes of information can be transmitted. . Each category has a range of allowed values, and for a particular piece of content, a particular category can have a single value or multiple values. Furthermore, a meta information group ("PICS label"
) May include deadline information. There is also a function that allows the PICS label to be applied to more than one part of the electronic content. Each PICS label in a particular portion of the electronic content can be independently added to or removed from the content.

For example, an image file with a PICS label that indicates in the “Rating Service” field that it contains a value-based rating label according to the “SafeSurf” rating system may be transmitted from the server. According to a preferred embodiment of the present invention, as it passes through the corporate proxy, the image file can be given a new updated category value on the PICS label to reflect local considerations of the "rating service" . Therefore, the client computer is
Display only new category values for labels. The HTTP protocol has been enhanced so that its request and response headers support PICS. The technical bodies that define other common application protocols, such as NNTP, are also now PIC
We are considering adding S support. As part of these protocols, a list of the desired PICS label types can be included in the request. PICS also
It defines an inquiry format for requesting PICS information of the central label office server.

The sample PICS label is (http: //the.content) exp “1997.0
7.09T08: 15-0500 "r (n4s3v2) PICS-1.
1 "http: //the.rating.service" label). Here, “n”, “s”, and “v” are transfer names of various meta information types, and values applicable to this content are 4 (for n), 3 (for s), and 2 (in the case of v). ID "ht
Only software that recognizes "tp: //the.rating.service" knows how to interpret these categories and values.

In a preferred embodiment, three types of PICS labels are introduced. The first type of PICS label, called a pre-fetched label (ie, a P-label), is used by server nodes to provide PHI. A second type of PICS label, called a request label (ie, an R label), is used to request a prefetch candidate object. A third type of PIC called a reference label (ie X label)
The S-label is used to provide feedback on the effectiveness of the prefetch based on which prefetch object is actually referenced.

The pre-fetched labels can include, but are not limited to, any combination of the following categories, including an identifier category, a probability category, a size category, and a caching category. Identifier category (that is, ID category)
Has a value that conveys the URL of a prefetch candidate. Probability category (ie, PRO
B category) has a value that conveys an estimate of the usefulness of prefetching. In the preferred embodiment, reference probability estimation is used. The reference probability indicates a probability that the prefetching candidate is referred to. The size category has a value that conveys the size or memory requirements of the prefetch candidate. The caching category has its first primary cache 294
Alternatively, it has a value that conveys information on an upstream proxy server that has a copy of the prefetch candidate object to any of its first prefetch caches 296.
In the preferred embodiment, only the closest superior proxy is identified.

The request label can include, but is not limited to, the following categories, including the identifier category (ie, ID category). The value of the identifier category conveys the URL of the object to be prefetched. In a preferred embodiment, if the (proxy or client) node wishes to pre-fetch suggested suggested pre-fetch candidate objects, the HTTP head request, in order to communicate the URL of the pre-fetched object, uses the request label And sent back to the content server 20 through the proxy hierarchy. When proxy server 24 receives a prefetch request from a lower hierarchical node, it processes the request if the object is in its primary or prefetch cache. Otherwise, forward the prefetch request to the next higher level proxy.

The reference labels may include, but are not limited to, the following categories, including the object identifier category (object ID category) and the reference side identifier category (reference side ID category). The object identifier category has a value that conveys the URL of the referenced prefetched object. The reference-side identification category has a value that transmits the URL of an object whose PHI includes the object specified by the object identifier category value.

In a preferred embodiment, when the pre-fetched object is actually referenced,
An HTTP head request is sent back to the content server 20 through the proxy hierarchy using the reference label to convey the URL of the referenced prefetch object and its referencing object. This type of HTTP head request is hereinafter referred to as a prefetch reference message. When the proxy server 24 receives the prefetched reference message from the lower hierarchical node, its first statistical table 293
And forward the request to the next higher-level proxy server.

FIG. 5 shows an example of the server logic 268 that updates the second statistical table 267 of the content server 20. The update of the second statistics table 267 is triggered by an object request from the client 22. The content server 20 preferably generates prefetch hint information to annotate or insert into the object header of the object before sending the requested object to the client 22.

First, at reference block 405, content server 20 waits for input. If, at decision block 410, the received input is an object request for object O, then at reference block 415 the statistics update unit 252 or the second update statistics routine 2
69 is called. At reference block 420, the generation unit 266 for generating the pre-fetch hint information 266 or pre-fetch generation routine is called. A detailed example of the generation unit 266 or PHI generation routine will be described later with reference to FIG. At reference block 425, for example, a PHI is inserted into the object header of the request object. Otherwise, at reference block 425, the prefetch hint information annotates the request object or request information. In a preferred embodiment, the PICS protocol can be used to annotate request objects. The PHI is captured in the P-label, where the identifier category value reflects the URL of the prefetch candidate. At block 430, the request object is returned to the requesting client or requesting proxy server. If, at decision block 435, the received input is a prefetch reference message (ie, an HTTP head request indicating that the prefetch object O was actually referenced with an X label), then at step 440, the statistics update means 252 or the second The update statistics routine 269 is called. For other types of input that are not the focus of the present invention, for example, FTP (File Transfer Protocol) requests, the appropriate miscellaneous handler can be invoked at reference block 445.

FIG. 6 shows how the statistics update unit 252 updates one of the statistics tables 261. The statistics update unit 252 includes a first update statistics routine 264 for updating the first statistics table 293, a second update statistics routine 269 for updating the second statistics table 267, or a first update statistics routine 264.
And both the second update statistics routine 269. The statistics table 261 is
Either the first statistics table 293 or the second statistics table 267, or the first
It refers to both the statistics table 293 and the second statistics table 267. The statistics update unit 252 first searches the referencing object for information. The request object is then placed on the referral object's access list or prefetch list.

First, at reference block 505, the statistics table 261 is updated to indicate that the object O has been referenced again by incrementing the first count. In other words, at block 505, the T count (0) is incremented by one. If the object O is not already in the statistics table 261, the parent object identifier is inserted into the statistics table 261, and the T count (0) is initialized to 1. If there is no empty slot in the statistics table 261, the longest unreferenced entry in the statistics table 261 is replaced with a new parent object identifier. Decision block 510
The requested object is checked for reference side information, for example, by checking the object header of the requested object. For example, in the HTTP protocol, a referencing object is provided in the header to indicate which object contains the HTTP link to the requesting object. If the referencing object (R) is in the header, at step 520, the statistics table 261 is examined to see if the object O is already on the access list of the object R. If so, at block 530, the second count (Rcount (0)) of the requested object is incremented by one. Otherwise, at block 525, the request object (object O) is inserted into the parent object (object R) access list and the second count (R count (0)) is initialized to one.

FIG. 7 shows an example of the generation unit 266 for generating the pre-fetch hint information in the pre-fetch hint information list. If the ratio of the second count to the first count exceeds the hint threshold, the pre-fetch hint information list includes the child object identifier of the child object. The pre-fetch hint information list represents a refined or filtered access list. That is, the generation unit 266 receives an input from the statistics table 261 and filters the input to obtain a pre-fetched hint information list.

The generation unit 266 for generating prefetch hint information represents a prefetch hint information information (PHI) generation routine 266, which begins at reference block 610. At reference block 610, the access list of the parent object (object O) is examined to see if it is empty (ie, whether A-list (O) is empty). If the access list is not empty, at reference block 620,
Let Cj be the next child object (child object identifier) identified in the access list of the parent object (object O).

At decision block 630, the reference probability of the child object Cj is compared with a hint threshold (TH), which is a pre-specified threshold (eg, 0.75). In a preferred embodiment, the reference probability of Cj is the ratio of the second count to the first count (R count (cj
) / T count (0)). This ratio is an estimate of the probability that a child object (object Cj) will be referenced after a parent object (object O) has been referenced. If, at decision block 630, the reference probability of Cj is greater than the hint threshold (TH), then at reference block 640, the child object identifier (Cj) is P
Included in the HI list. In other words, if the ratio exceeds the hint threshold, the child object identifier and the related information are put on the pre-fetched hint information list.

Generation unit 266 may require a software subroutine or software component to perform the comparison at reference block 630. In particular, the generating unit 266 for generating the pre-fetch hint information in the pre-fetch hint information list includes a calculating means for calculating the reference probability and a comparing means for comparing the reference probability with the hint threshold value. Can be.

In a preferred embodiment, using the PICS protocol, the PHI information is stored in the P label, where the URL of the child object (Cj) is received as an identifier category (ID category value), its reference probability and size Are received as probability category (PROB) and size category values, respectively. Decision block 6
At 45, if the access list of object O has not been completely scanned, reference block 620 is re-executed.

Various alternative criteria can be used to select pre-fetch candidates from the statistics table 261. For example, a moving average of reference probabilities in a past section can be used as a reference. Further, the selection procedure of the prefetching candidate can be recursive. When the child object (object Cj) is selected as a candidate for pre-fetching the parent object (object O), an arbitrary child object on the access list of the child object (object Cj) is evaluated. Can determine whether to include in the pre-fetch list of the parent object (object O). At decision block 630, the comparison or evaluation unit now tests whether the product of the R count (Cj) / T count (0) and the R count (Cj1) / T count (Cj) is greater than a hint threshold (TH). can do.

FIG. 8 shows an example of the operation of the client logic 244. FIG. 8 shows that once the object is requested, the client object request handler is invoked. Otherwise, the prefetch handler 241 can process the unrequested object but needs to store it in the second prefetch cache 249 of the client 22.

First, at block 710, the client 22 waits for input. Decision block 72
At 0, if the input received is an object, the second prefetch handler 241 is called at reference block 730. At block 740, the client object handler is called to return the object to the requesting application at client 22. At decision block 750, if the received input is an object request, then at reference block 760 the client object request handler 247 is called. For other types of input that are not the focus of the present invention (eg, a push request), the appropriate miscellaneous handler can be invoked at reference block 770.

FIG. 9 illustrates the first prefetch handler 248 or the second prefetch handler 241.
Or an example of the operation of the pre-fetch unit 250 including both the first pre-fetch handler 248 and the second pre-fetch handler 241. In summary, the pre-fetch unit 250 uses the pre-fetch list generated by the generation unit 266 of FIG. It is determined whether or not storage is guaranteed. The pre-fetch unit 250 calculates a pre-fetch value for each pre-fetch object candidate or child object identifier on the pre-fetch hint information list. The prefetch value takes into account the access probability of the child object and the access cost of storing the child object. Starting from the child object having the highest threshold, all child objects are stored until one free space of the prefetch cache 298 is filled.

First, at block 805, let L be the PHI list of prefetch candidate objects. The pre-extraction candidate object can be represented by a child object identifier on the pre-extraction hint information list. Prefetchable candidates that appear in the local cache (ie, prefetch cache or main cache) are removed from L. At block 810, a prefetch value (PV) for each object in L is calculated. In a preferred embodiment, the PV value is defined as the reference probability multiplied by the access cost. The access cost is estimated based on the closest caching location of the prefetch candidate object or child object identified on the PHI list. As a practical matter, child objects are preferably identified on the PHI list by a child object identifier, rather than as a whole child object, to save storage space. Optimally, the reference probability is obtained from the probability category (PROB category) value, and the closest caching position is obtained from the caching category value of the PICS label. At block 820, the prefetch candidate object having the highest PV value is selected. At reference block 830, the amount of prefetch buffer space available to accommodate the candidate prefetch object (Oj) is determined. This space is currently unused or occupied by objects with lower PV values, which are only a fraction (eg, one half) of the PV values of the current pre-fetched object candidate (Oj) to be evaluated. Is either. Storage space occupied by objects marked as removable in one of the prefetch caches 298 is considered to be currently unused. At decision block 840, one of the prefetch caches 298 is examined,
It is checked whether there is enough space for Oj. In some cases, at reference block 850, object Oj is prefetched from one of prefetch cache 298. At block 860, Oj is removed from the PHI list (L). If L is not empty at reference block 870, the steps associated with reference block 820 are re-executed.

In the preferred embodiment, each object is pre-fetched one at a time,
Multiple objects can be pre-fetched together in a single request. At reference block 850, Oj can be placed on the prefetch list. Block 870
, If L is empty, then a prefetch request is issued with the entire prefetch list. Further, even if there is not enough storage space in the prefetch cache, the pending prefetch list can be maintained, so that additional prefetch can be initiated when storage locations become available. .

By using the HTTP protocol, a communication session can be kept open during pre-fetch to reduce communication overhead. There is no need to re-establish the session for each prefetch.

FIG. 10 shows an example of the operation of the client object request handler 247. In summary, the client object request handler 247 first searches the second primary cache 246 and then the second prefetch cache 249.
If the object is not in the second main cache 246 or the second prefetch cache 249, the object request is sent to the server. If the object is in the second prefetch cache 249, a prefetch reference message is sent to one of the servers to indicate that the prefetch object was actually referenced.
Then, the second prefetch handler 241 and the client object handler are used.

First, at decision block 905, the requested object O is
If found at, the second prefetch handler 241 is called at reference block 925 to initiate the appropriate prefetch operation based on the PHI associated with object O. At reference block 930, the client object handler is called to return the requested object to the requesting application at the client. If the requested object O is found in the second prefetch cache 249 at decision block 910, block 9 is executed so that its second statistics table 267 can be updated to capture the actual reference of the prefetch object.
At 20, a prefetch reference message is sent to the content server 20. In effect, the prefetch reference message is passed through the proxy hierarchy to the content server 20.
So that the second statistics table of each intermediate proxy is also updated. If the object is not in the second prefetch cache 249 at block 910, then at reference block 915, the object request is sent to the content server 20 through the proxy hierarchy.

The prefetch reference message can be batched. That is, the prefetched reference list can be maintained, and the compound prefetched reference message is periodically sent to the upper server in order to confirm the references of all objects on the prefetched reference list.

FIG. 11 shows an example of the operation of the proxy server logic 295. In summary, the input is stored depending on whether the input is an object request, an object transfer, or a prefetch reference message. Depending on the input, three types of handlers or software routines are available: a proxy object request handler 283, a proxy object handler 284, and a prefetch reference message handler 285.

First, at reference block 1005, the proxy server 24 waits for input. If, at decision block 1010, the received input is an object request for object O, then at block 1015 the proxy server 24 checks whether there is already a pending request for that object O. The pending request may be due to a prefetch request for that object O sent by this proxy server 24, or a previous request to object O originating from another proxy server lower down in the proxy server hierarchy. It can be. If there is no pending request for object O in reference block 1015, then proxy object request handler 283 is called in block 1020. Decision block 10
At 25, if the input received is an object (from an upstream proxy), then at reference block 1030, the proxy object handler 284 is called. Otherwise, if the input received at block 1040 is a prefetch reference message, then at block 1045 the prefetch reference message handler 285
Is called. For other types of input that are not the focus of the present invention (such as an FTP request), the appropriate miscellaneous handler can be invoked at block 1060.

FIG. 12 shows an example of the operation of the proxy object request handler 283. In summary, if an object is in the prefetch cache, the object
Returned or sent to the requesting client or requesting proxy server.
Thereafter, if the request object was a prefetch request, the request object is marked as removable from the first prefetch cache 296 of the proxy server 24. If a prefetch request was not provided to the request object, a prefetch reference message is sent to the content server 20. However, the prefetch reference message is sent because the request object is in the first prefetch cache 2
Only when sent from 96. If the requested object is not in the first prefetch cache 296 or the first main cache 294, the object request is sent to another proxy server 24 or content server 20. The request object found in the first main cache 294 is returned or sent to the requesting client or proxy server 24, and if the request was not a prefetch request, the first statistics table 293 is updated.

Initially, at decision block 1105, if the requested object O is found in the prefetch cache, at block 1130, the object O is returned to the requesting node. If the request is a prefetch request at decision block 1135, block 1150
The object O of the first prefetch cache 296 is marked as deleteable. Deleting an object from the first prefetch cache 296 provides space for future prefetch as the prefetch object O has now been prefetched by some nodes below the proxy server 24. Otherwise, at reference block 1140, the cache manager is called and object O is actually referenced and object O is now in the first prefetch cache 2
From 96, it is moved to the first main cache 294. At block 1160, a prefetch reference message is sent to the content server 20 through the proxy hierarchy server. At block 1170, the update statistics routine is called and the local first statistics table 293 is updated to capture the actual reference of the prefetched object O. If the object O is not found in the first primary cache 294 at decision block 1110, the object request is sent to the content server 20 through the proxy server hierarchy at reference block 1125. At decision block 1115, if the object request is a pre-fetch request, then at block 1120 the first update statistics routine 264 is called to make the local first
The statistics table 293 is updated. At block 1180, object O is returned to the requesting node.

FIG. 13 shows an example of the operation of the prefetch reference message handler 285. Prefetch reference message handler 285 first checks to see if the object is in first prefetch cache 296. If the object is in the first prefetch cache 296, the cache manager routine is called before forwarding the prefetch reference message. If the object is not in the first prefetch cache 296, a prefetch reference message is forwarded.

First, in step 1205, the first update statistics routine 264 is called, and the local first statistics table 293 is updated to capture the actual reference of the prefetched object (O). At decision block 1210, if the object O is in the first prefetch cache 296, the cache manager is called at reference block 1215 and the object O is first referenced because the object was actually referenced.
The prefetch cache 296 is moved to the first main cache 294. At block 1220, the prefetch reference message is forwarded to the next higher level proxy.

FIG. 14 shows an example of the operation of the proxy object handler 284. The proxy object handler 284 first starts with the prefetch update unit 291
Alternatively, the prefetching hint information update routine is called uniformly without exception. Prefetched object requests are handled differently depending on whether the prefetched object was requested by a subordinate server. If a pre-fetched object was requested by a child server, the object is returned to that child server. Otherwise, the prefetch cache manager is called. If a prefetch object is not requested, the object is returned to the requesting proxy server 24 or client, and a first update statistics routine 264 is called.

First, at reference block 1305, to update the PHI of object O,
The PHI update routine or the prefetch update unit 291 is called. If the object received at reference block 1310 is not a pre-fetched object,
In step 1315, the first prefetch handler 248 is called. At reference block 1320, the object is returned to the requested node (lower hierarchy). PIC
In a preferred embodiment using the S protocol, the caching category value in the P label is updated if the corresponding prefetch candidate (specified by the identifier category value) is prefetched by this proxy server 24. . At block 1325, the cache manager is called to store the object in the main cache. At reference block 1355, the update statistics routine is called. If, at block 1330, the received pre-fetched object is requested by a lower node in the hierarchy, the object is returned to the requestor at reference block 1350. If not, it is requested by the current node and the prefetch cache manager is called at reference block 1340 to store the object in the prefetch cache by making the necessary substitutions to make room for the object. You.

FIG. 15 shows an example of the prefetch update unit 291 or the PHI update routine.
The PHI update routine compares the request object's local count to the threshold count. If the requested objects exceed a first minimum threshold, they undergo a second minimum threshold test. The threshold can be set based on the test results. If the local count is less than the first minimum threshold, the original pre-fetch hint information list is maintained.
If the local count is greater than the first threshold but not greater than the second threshold, the prefetch hint information list is changed.

First, at decision block 1405, if the local count (T count) of object (O) in local first statistics table 293 is greater than the first minimum threshold (CTM), the original PHI in the header of object O is maintained Is done. This is simply due to the fact that there are not enough object references in the lower tier of the proxy server 24 to guarantee the update operation. If, at decision block 1415, the local count (T count) of the object (O) in the local statistics table 261 is greater than a second minimum threshold (TCH), the original PHI in the object header is ignored and at block 1420, To generate a PHI based on the local first statistics table 293, a PHI generation routine is called. This is the proxy server 24
There are enough references to objects below the hierarchy to generate a PHI based on local actions. Otherwise, the PHI at reference block 1425
The change routine is called to change the probability category value by taking a local estimate of the reference probability for each object in the PHI list and averaging it in the header (set by the upper layer).

The prefetch scheme works even if some of the proxy servers 24 do not maintain the first statistics table 293 and track local reference patterns. The untracked proxy server simply accepts the probability (ie, PROB) category value in the PHI passed down from the upstream proxy server or content server 20.

The prefetch scheme also works in a heterogeneous proxy server environment where some of the proxy servers are conventional proxy servers that do not understand the prefetch protocol and do not participate in cooperation. These traditional proxy servers are
The first statistical table 29 is not interpreted without starting the prefetching operation by interpreting the header.
No 3 is maintained. A conventional proxy server either passes meta data (eg, a PICS label in the header) along with the object to the next level proxy server, or if it decides to cache the object locally, It simply stores the meta data with the object.

The present invention can be applied to any network or client-server environment, even if there is no proxy hierarchy. The object sender node can track the reference pattern, derive the PHI, and include it in the object header. In addition, the object receiver node can store the PHI with the object in its cache. After referencing the object, a prefetch operation can be initiated based on the stored PHI and local factors such as cache size and access costs as described above.

Storing the PHI with the objects to be cached provides additional opportunities for prefetching. Using events such as more storage locations becoming available, additional references occurring, or receiving prefetched objects,
By examining the PHI of the cached object, an additional prefetch can be triggered. The prefetch operation is not constrained at the moment when the object is requested from the server, as in the case of the activation method at the sender.

Although the preferred embodiment has described the general pre-fetch policy for web and proxy servers, those skilled in the art will appreciate that the present invention can be used in any kind of situation where pre-fetched objects have similar characteristics. It will be understood that this is also applicable and is not necessarily limited to Internet or WWW applications only.

Furthermore, the preferred embodiment of the present invention is the case of cooperative prefetching between parent and child proxies in a hierarchy, but it is easily adapted to include cooperation between sibling proxies. Can be. For example, if a higher-level proxy does not cache the request object or prefetch candidate, the proxy can also query a sibling proxy.

[Brief description of the drawings]

FIG. 1 Content server (web site), network (Internet),
FIG. 2 is a block diagram illustrating a proxy server hierarchy and clients.

2 is another block diagram illustrating the content server, client, and proxy server in more detail than FIG.

FIG. 3 is a flowchart illustrating a method for dynamically pre-fetching data from a content server through a hierarchy of proxy servers.

FIG. 4 is a diagram illustrating an example of a statistics table.

FIG. 5 is a flowchart showing the operation of the content server.

FIG. 6 is a flowchart showing updating of statistics.

FIG. 7 is a flowchart showing generation of advance extraction hint information.

FIG. 8 is a flowchart showing the operation of the client.

FIG. 9 is a flowchart showing advance removal.

FIG. 10 is a flowchart showing the operation of a client object request handler.

FIG. 11 is a flowchart showing the operation of the proxy server.

FIG. 12 is a flowchart showing the operation of the proxy object request handler.

FIG. 13 is a flowchart showing the operation of the prefetch reference message handler.

FIG. 14 is a flowchart showing the operation of the proxy object handler.

FIG. 15 is a flowchart showing the operation of a prefetching hint information updating routine.

──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Yu, Philip, Sea Lang United States 10514 Chapaqua Stornoway, New York 18 F term (reference) 5B075 KK04 KK07 ND20 ND36 NK02 PQ02 PQ32 PQ42 PR03 PR08 QP10 QT03 UU40 5B082 FA03 FA11 GC04 5K030 HA HC01 JT02 KA01 KA06 KA07 [Continued Summary] The final decision to prefetch selection information is based on prefetch hint information and prefetch hint values. Prefetch hint values are calculated based on local node specific factors such as storage constraints and access times.

Claims (35)

[Claims] 1. One or more clients (22) associated with a content server.
A communication network (25) and one or more proxy servers (24)
Prior to retrieving the selection information from the content server (20) so that the selection information can be easily accessed via the method, requesting the information from at least one of the proxy servers (100) Establishing pre-fetch hint information for the client based on a past search or access of request information by any of the clients (
102, 104) and a step of adding the pre-fetched hint information to the request information as an annotation (105).
) And pre-fetching selection information from any of the servers based on pre-fetch hint information and request information (108). 2. The method of claim 1, wherein the request information includes a request object and the selection information includes a pre-fetch candidate object. 3. The method of claim 1, wherein the step of establishing prefetch hint information comprises: tracking a data reference pattern associated with the client; and organizing the data reference pattern into a group of object identifiers. Forming a group according to the relationship between the request object and the pre-fetch candidate object; and ensuring that the pre-fetch candidate object is pre-fetched as selection information between the pre-fetch candidate object and the request object. Determining which of the groups has a sufficient relationship. 4. The step of tracking and organizing data reference patterns comprises: establishing a database for storing reference data patterns in a proxy server; and arranging fields in each of the databases. Step, wherein the fields include an object identifier and an object count, wherein the object identifier distinguishes request objects from each other, and wherein the object count indicates a number of requests for the request object observed by one of the servers. 4. The method of claim 3, comprising: associating an access list with the database, wherein each of the access lists includes one or more prefetch candidate objects associated with each request object. of Law. 5. The step of tracking and organizing a data reference pattern comprises: establishing a database defining a relationship between a request object and a prefetch candidate object; and tracking access statistics for the request object and the prefetch candidate object. The access statistics include a parent object access frequency of the request object and a child object access frequency of the prefetch candidate object, wherein the parent object access frequency is such that a particular one of the request objects is The child object access frequency represents the number of times one or more of the prefetch candidate objects was actually requested immediately after the particular one access. Further comprising the steps, the method according to claim 3 or 4. 6. The method of claim 1, further comprising: selecting a prefetch list of prefetch candidate objects based on a ratio of the child object access frequency to a parent object access frequency associated with the child object access frequency. The method of claim 5, wherein the pre-fetch list includes only pre-fetch candidate objects having a ratio that exceeds a threshold probability. 7. The method according to claim 1, wherein the step of pre-retrieving selection information comprises: pre-retrieving a subset of pre-retrieve candidate objects received at one of said proxy servers or one of said clients. Selecting the subset of the pre-fetched candidate objects as selection information based on available storage capacity at any one time. 8. The method of claim 1, further comprising: retrieving a subset of prefetched candidate objects received at one of the proxy servers or one of the clients; Selecting a subset of the pre-fetch candidate objects as selection information based on the access time in one of the plurality of candidate objects. 9. The proxy object, wherein the selected object and the request object include Internet data, wherein the request object references the selected object via a hypertext transfer protocol (HTTP) link associated with the Internet data, and wherein the proxy object includes: 9. The server of claim 2, wherein the server identifies the relationship of the selected object to the request object through the request object's Universal Resource Locator (URL) HTTP protocol stored in the request object's object header. the method of. 10. The step of annotating the request information comprises placing the prefetch hint information on a prefetch label having a platform in an Internet Content Selection (PICS) format or a label protocol compatible with the PICS format. The method according to any one of claims 2 to 9, comprising a step. 11. The internet content selection (PICS) format platform includes an object identifier category and a referrer identifier category, and wherein the method includes the step of identifying the referrer identifier to identify a pre-fetch candidate object that was actually referenced. The method of claim 10, further comprising returning the category to a content server. 12. The method according to claim 12, wherein the step of establishing the pre-fetch hint information comprises the step of identifying a data reference pattern of the client based on a past search or previous access to request information, wherein the data reference pattern is Selected to represent a general trend of past searches or accesses by various of the clients associated with one of the proxy servers; and defining prefetch hint information from the data reference pattern. The method according to claim 1 or 2, comprising the steps of: 13. There is a hierarchy of the proxy server, wherein the hierarchy includes a sending node and a receiving node, wherein the method comprises: transmitting prefetch hint information from the sending node to the receiving node; Changing the extraction hint information from an initial state to a final state, wherein the initial state is determined by a transmitting node, the final state is determined by the receiving node, and the pre-extraction hint information of the final information is final pre-extraction. 13. The method according to any one of the preceding claims, further comprising: a step designated as a decision. 14. The method of claim 13, wherein the step of pre-retrieving the selection information is triggered by a receiving node. 15. The step of organizing proxy servers into a hierarchy of servers, the hierarchy having a plurality of levels of proxy servers, wherein the plurality of levels are in proximity to a network of each of the proxy servers. Determining the level closest to the net to the highest level, and interpreting and updating the prefetch hint information transmitted from the proxy server to the client via the hierarchy. 13. The method according to any one of the preceding claims, comprising: 16. The proxy server of claim 15, wherein the proxy server updates the prefetch hint information if there is a local reference to the request information and if the local level server count exceeds a minimum threshold count. Method. 17. The method according to claim 15, wherein the prefetch hint information is updated based on a caching state of a proxy layer. 18. A data processing system for prefetching selection information from a content server (20) via a network (25), the system comprising one or more proxy servers (24) associated with the network. A data processing system comprising: a client for communicating with one or more of a content server and a proxy server via a network; means for requesting information from at least one of the proxy servers; Means for establishing prefetching hint information for the client based on a past search or access of the request information by any of: a means for adding the prefetching hint information as an annotation to the request information; Based on the pre-fetching hint information and request information, Data processing system including a means for extracting in advance a selection information from any of the bus. 19. The request information includes a request object, the selection information includes a prefetch candidate object, and each of the proxy servers and at least one of the clients has at least one between a main cache and a prefetch cache. 19. The processing system of claim 18, comprising a memory with logical separation, wherein prefetch selection information is stored in a prefetch cache. 20. The prefetching means for prefetching an object and storing the prefetched object in a prefetch cache based on a priority configuration, wherein the priority configuration is the highest in free memory of the prefetch cache. 2. The method of claim 1, wherein an object having a prefetched value is first stored, wherein the prefetched value is based on individual or collective reference probabilities to the prefetched object by one or more clients.
10. The processing system according to 9. 21. The prefetched value may be an estimated access probability that a prefetched candidate object on the prefetched hint information list is accessed by a particular one of the clients, and a selected object of the prefetched candidate object. 21. The processing system of claim 20, wherein the cost of accessing the object is based on a required storage capacity for storing the object. 22. The prefetching means includes a pre-read handler including calculating means for calculating the prefetched value and evaluating means for evaluating the prefetched value calculated by the calculating means. The calculating means calculates a pre-fetched value of each pre-fetched object identifier on the pre-fetched hint information list, the evaluating means evaluates pre-fetched values of a plurality of candidate pre-fetched objects, and pre-fetched hint information 22. The processing system according to claim 20 or 21, wherein priority is given to caching of candidate prefetch objects associated with the highest prefetch value in the list. 23. The method according to claim 23, wherein the pre-fetching means comprises a first server in at least one of the proxy servers.
A prefetch handler and a second in at least one of the clients
23. The processing system of claim 22, comprising a prefetch handler. 24. The system further comprising a client object request handler stored in a memory of one of the clients, the client object request handler pre-fetching and pre-fetching the request object before requesting the request object from a proxy server or a network. Looking up the request object in the cache, and if the object is in the prefetch cache, the client object request handler sends a prefetch reference message to the server indicating that the prefetch object was actually accessed by the client. 20. Notifying a content server or a proxy server by generating.
4. The processing system according to claim 3. 25. A prefetch reference message handler for forwarding a prefetch reference message after invoking a cache manager routine if at least one of the proxy servers has a request object in a prefetch cache. The processing system according to any one of claims 19 to 24, comprising: 26. A content server or proxy server for use in a data processing system according to any one of claims 18 to 25, wherein said request information comprises a request object and said selection information is prefetched. A statistics table storing candidate object objects, the server storing a parent object identifier and an associated child object identifier, the server including a first count of the parent object representing a frequency with which the parent object is referenced by the client; A statistics table including a second count of the child object indicating a frequency of referring to the corresponding child object after the selected object of the objects is referred to; and generating a prefetch hint information list based on the information of the statistics table. Generating means for performing Server. 27. The generating means receives an input of a first count and a corresponding second count to derive a pre-fetched hint information list, and the generating means generates the pre-fetched hint information list for each of the parent object identifiers with respect to the first count. Means for calculating a ratio of the second count, and comparing means for comparing the ratio with a hint threshold, wherein the comparing means, when the ratio exceeds the hint threshold, displays on the pre-fetched hint information list. 27. The server of claim 26, which groups or associates child objects and associated parent objects. 28. The generating means accepts input of a first count, a second count, a child object identifier, and a parent object identifier from a statistics table, wherein the generating means determines a relationship between the parent object and their associated child objects. Evaluating means for evaluating the access relationship between the prefetching hint information if the child object is referred to immediately at a suitable frequency immediately before the client refers to the corresponding parent object. 27. The server of claim 26, wherein the server places the child object identifier on the list. 29. The server of claim 28, wherein said generating means further comprises calculating means for calculating said appropriate frequency based on a ratio of child access to corresponding parent access. 30. A statistic updating means for updating statistics based on actually requested child objects, said statistic updating means first searching for an identity of a referencing object of said parent objects, 30. The method of claim 26, wherein the referencing or referencing object is referenced before the actually requested child object, and the actually requested child object updates its corresponding second count in the statistics table. The server according to claim 1. 31. The server according to claim 30, wherein the statistics updating unit updates the statistics table when receiving a prefetch reference message from a prefetch reference message handler. 32. The server, wherein the server is a proxy server, further comprising prefetch update means for receiving and updating a prefetch hint information list, wherein the prefetch update means comprises a first count and a first local threshold count. The prefetch update unit calls the generation unit for each child object identifier whose first count exceeds the first local threshold count, and the prefetch update unit checks that the first count is equal to the first local threshold value. 32. The server according to any one of claims 26 to 31, wherein an original pre-fetch hint information list is maintained if the threshold count is not exceeded. 33. The prefetch update means compares the first count with first and second local threshold counts, and (i) if the first count does not exceed the first local threshold count, the original prefetch hint. Maintaining the information list, (ii) modifying the original prefetch hint information list if the first count exceeds the first local threshold count but does not exceed the second local threshold count, and (iii) the first count is 33. The server of claim 32, exchanging the original pre-fetch hint information list if both the first and second local threshold counts are exceeded. 34. A proxy object handler for returning the request object to a subordinate proxy server or client that has requested the request object, the proxy object handler comprising: 34. The server according to claim 32 or 33, wherein the server returns the request object and updates the statistics table accordingly. 35. A prefetch cache manager for managing a prefetch cache, the proxy object handler further comprising: a requesting client or proxy proxy if the object is requested according to a prefetch request. Returning the request object from the prefetch cache to the server and marking the object as removable from the prefetch cache, the proxy object handler returns an object if the request object is not in the main cache or the prefetch cache. Sends the request to another server, and the proxy object handler determines if the request object is not associated with a prefetch request and if the request object is not in the prefetch cache or main cache. If that updates the statistics table accordingly, server of claim 34.